流行病學(xué)(英文)-醫(yī)藥類課件

Case-controlstudiesThisversionismadeforbilingualteaching.Case-controlstudyisanessentialresearchdesignofEpidemiology,whichinvolvesidentifyingpatientswhohavetheoutcomeofinterest(cases)andcontrolpatientswhodonothavethatsameoutcome,andlookingbacktoseeiftheyhadtheexposureofinterest.Theexposurecouldbesomeenvironmentalfactor,abehaviouralfactor,orexposuretoadrugorothertherapeuticintervention.SelectStudyDesigntoMatchtheResearchGoalsCase-ControlStudiesIntroductionMatchingInvestigateExampleDesignofCase-ControlStudiesDatacollectionandanalysisBiasStrengthsandWeaknessesSeveralimportantfeaturesIntroductionHistoricalPerspectiveDefinitionTypesofDesignUniquecontributionofepidemiologytotherepertoireofclinicalresearchdesignsFirstcase-controlstudyperformedinlate1950sDollandHill’sstudyoflungcancerandsmokingbehavioramongphysiciansJeromeCornfield’sclassicdescriptionof“RetrospectiveStudies”Newstatisticaltoolsweredevelopedtoanalyzethestudydesign-logisticregressionHistoricalPerspectiveIntroductionDefinition

Acase-controlstudyisadesigninwhichindividualswithaneventorconditionofinterest,CASES,areidentifiedandthencomparedwithregardtooneormoreexposurestoindividualswithouttheeventorconditionofinterest,CONTROLS.Case-controlinvestigationstypicallyaredesignedtoassesstheassociationbetweenoccurrenceofdiseaseandanexposuresuspectedofcausing(orpreventing)thatdisease.IntroductionabcdCasesControlsDirectionofinquiryExposedExposedUnexposedUnexposeda/(a+c)b/(b+d)IntroductionTypesFamilyofepidemiologicalstudydesignsTraditionalcase-controldesignCase-controlstudieswithincohortsNestedcase-controlstudydesignCase-cohortstudydesignCase-parentstudydesignCase-onlystudydesignIntroductionMatchingSummarizeTypesProblemswithMatching

MatchingisdefinedastheprocessofselectingcontrolssothattheyresemblethecaseswithregardtocertaincharacteristicsThegoalofmatchingistocreatesimilardistributionsbetweencasesandcontrolswithregardtocertaincharacteristicsMatchingcanbeusedtoAdjustforpotentialconfoundingfactorsIncreaseprecisionofestimateMatchingSummarizeIndividuallevelmatchingForeachcaseinthestudy,oneormorecontrolsareselectedwithidentical(similar)characteristicsasthecaseFrequency,orgroup,matchingSelectcontrolssothattheproportionwithacertaincharacteristicisidenticaltotheproportionofcaseswiththatcharacteristicMatchingTypes

DifficultandexpensiveCannotevaluatetheeffectofcontrolledvariablesMaylimittheabilitytocontrolforothervariablesOvermatchingControlsresemblecasesintermsofknownandunknowncharacteristics,someofwhichmaybeassociatedwiththediseaseProblemswithMatchingMatchingInvestigateExampletheassociationbetweenoccurrenceofEosinophilia-myalgiasyndrome(EMS)andIngestionofL-tryptophan.BackgroundconductresultsInvestigateExampleBackgroundEMSwasfirstrecognizedinOctober1989,itoccurspredominantlyinwomenandisrelativelyrare.whenastutephysiciansdeterminedthatthreepeoplewithunexplainedmyalgiasandeosinophiliahadconsumedL-tryptophan.Promptresponsebyhealthdepartmentsquicklyledtocase-controlstudies,theresultsofwhichsuggestedthatingestionofL-tryptophanwasthecauseofEMS.TheCentersforDiseaseControlandPrevention(CDC)conductedaseriesofcase-controlstudiesin1989and1990.OneofthestudiesconductedinMinnesota,Researchersselected63casesubjectsofEMSinthemetropolitanareaofMinneapolis-St.Paul.Researchersrandomlyselected5188controlsubjectsinthesamearea.ResearchersinterviewedsubjectsandaskedabortpotentialriskfactorsandabouttheiruseofL-tryptophan.InvestigateExampleConductL-tryptophanwastakensignificantlymorefrequentlybycasesthanbycontrols—61of63casesubjects(97%),butonly101of5188controlsubjects(2%).L-Tryptophan-containingproductsweretakenoffthemarketinNovember1989,In1990,aftertherecallofL-tryptophan,thenumberofreportedcasesfelltonearzero.InvestigateExampleResultsSelectionofCasesDevelopacasedefinitionthenidentifynewcaseswithinaspecifiedtimeperiodSelectionofControlsThesampleofcontrolsshouldhavethesameprevalenceofexposureasthesourcepopulationofunaffectedpersons.DeterminationofExposureDesignSelectionofCasesDesignSourcesofcasesSpeciesofcasesSomethingimportantSelectionofCases

SourcesofCasesSourcesofCasesHospitalorclinicBecauseriskfactorsmayresultfromreferralpatternstospecifichospitals,multiplehospitals/clinicsoftenchosenReferralofmoreillpatientstohospitals,especiallytertiarycarecentersPopulation-basedorcommunityNewcasesreportedtohealthdepartments,registries,hospitalrecorddepartments,etc.CasescannotbeselectedbasedonknownorunknownassociationwithexposureofinterestDesignSelectionofCases

SpeciesofCasesNewlydiagnosedorincidentcasesPreviouslyexistingorprevalentcasesIncidentcasespreferredoverprevalentcasesinmostsettingsIfprevalentcaseschosen,thenriskfactorsidentifiedfordiseasemaybethoserelatedmoretosurvivalwithdiseasethandiseaseoccurrence.Survivorshipbiasalsotrueforincidentcases,butminimizedDesignSpecifythedefinitionofacaseThecriteriashouldminimizethelikelihoodthatanaffectedperson(truecase)ismissed(i.e,thecriteriamustbesensitive).Anonaffectedpersonisfalselyclassifiedasacase(i.e,thecriteriamustbespecific).Design

SelectionofCases

SomethingImportantSelectionofControlsSourcesofcontrolsMultiplecontrolsSomethingimportantDesignHospitalcontrolgroup

Hospitalizedpatients,bestifchosenfromthesamehospitalascasesinordertocontrolforunknownreferencepopulationSelectfromallpatientsadmittedtothehospitalSelectfromspecificdiagnosisDesignSelectionofControls

SourcesofControls(1)SelectionofControls

SourcesofControls(2)CommunitycontrolgroupProbabilitysamplebest,butnotoftenpracticalSelectfromschoolrosters,insurancecompanies,etc.NeighborsofcasesRandomdigitdialingBestfriendDesignSelectionofControls

MultipleControlsControlsofthesametypeMayimproveprecisionofthemeasureofassociationPrecisionrarelyimprovedwithmorethan5controlspercaseControlsofdifferenttypesHospitalcontrolsandcommunitycontrolspercaseDesignControlscannotbeselectedbasedonknownorunknownassociationwithexposure(s)orriskfactorsofinterestDesignSelectionofControls

SomethingImportantExposureSomethingimportant

DeterminationofExposureDesign

DeterminationofExposure

ExposureExposureisdeterminedina‘retrospective’manner,thatisonemustlookbackintimetoassessexposurestatusbeforeapersonbecameacase.Eachindividual’spriorexposuretotheriskfactorofinterestOtherexposuresDesign

DeterminationofExposure

SomethingImportant(1)CasesandcontrolsmustbeassessedforexposureinthesamewayInterviewsshouldbestandardized,monitored,andconductedbytrainedinterviewers.DesignExposuremustbemeasuredinablindedmannerDatacollectorsmustbeunawareofwhethersubjectisacaseorcontrolDatacollectorsshouldbeunawareofthestudyhypothesisDesign

DeterminationofExposure

SomethingImportant(2)DatacollectionandanalysisCollectionofDataAnalysisofDataORUnmatchedanalysisMatchedanalysisAnalyticStrategy

CollectionofDataInterviewsandquestionnairesInformationconcerningriskfactorsmayalsobeobtainedfrommedical,occupational,orotherrecords.Datacollectionandanalysis

AnalysisofDataDatacollectionandanalysisUnexposed-cExposed-aPopulationatRiskExposed-bCases

ControlsUnexposed-dThepowerofthestudydesignliesinthesymmetryoftheOR.ORistheoddsofexposuregivendiseasedividedbytheoddsofexposuregivennodisease.Rememberthattheoddsofexposureamongcasescomparedwithcontrolsisthesameastheoddsofdiseaseamongexposedandunexposed.Datacollectionandanalysis

OddsRatioExposedUnexposedTotalCasesControlsacbda+bc+dTotala+cb+da+b+c+dUnmatchedanalysis

DatacollectionandanalysisUnmatchedanalysisDatacollectionandanalysisControlExposedControlUnexposedTotalCaseexposedaba+bCaseunexposedcdc+dTotala+cb+da+b+c+d

Case-controlpairsthatsharethesameexposurestatusdonotcontributetotheestimateofrisk.DatacollectionandanalysisMatchedanalysisMatchedanalysisDatacollectionandanalysis

AnalyticStrategyAssessrelationship/associationbetweenExposureandindependentvariablesCase/ControlstatusandindependentvariablesCalculatecrude,orunadjusted,ORforexposure-caseassociationMatchedanalysisrequiredformatchedstudiesDatacollectionandanalysis

AnalyticStrategyStratifiedanalysisCalculatestratum-specificORsforexposure-caserelationshipDeterminepresenceofconfoundingandinteractionLogisticregressionanalysisRegressiontechniqueusedtoadjustforconfoundingandinteractionSpeciallogisticmodelappliedinmatchedstudiesDatacollectionandanalysisBiasIntroductionSelectionbiasInformationbiasConfounding

IntroductionCase-controlstudiesaresubjecttobiasandconfounding,bothwilldistorttheresultsofthestudyBiasisdefinedasthedeviationofresults,orinferences,fromthetruth,orprocessesleadingtosuchdeviation.Thereareabout75differenttypesofbiasnowidentifiedinpublishedcase-controlstudiesBiasSelectionBiasFeaturesTypesBiasFeatures(1)Selectionbiasreflectssystematicerrorsthatarisefromthewayinwhichsubjectsareselected.Ifthepriorexposureofthecasesstudieddiffersfromthatofallcasesarisingfromthesourcepopulation—orifpriorexposureofcontrolsdiffersfromthatofpersonsinthesourcepopulationwithoutthediseaseorinterest—selectionbiasmaybepresent.BiasFeatures(2)Preferentialdiagnosisofexposedcasesmayleadtoselectionbias.Lowparticipationmayleadtoselectionbias.Errorsinsamplingcontrolsfromthesourcepopulationcanalsocreateselectionbias.BiasTypesAdmissionratebiasPrevalence-incidencebiasDetectionsignalbiasTimeeffectbiasBiasInformationBiasAdistortioninmeasuringexposureoroutcomedatathatresultsindifferentquality(i.e.,accuracyorreliability)orfrequenc